Cant help myself but to revive this thread after all these years.
Those who are reading this must either have a broken SD135/HD220, or got a bit annoyed with the progression from a positive "dimensionality instead of paper cut outs and fixed images, roundness without bloat or hyper inflation of images " to a negative "excessive overemphasis in the upper octaves. Overall, it sounds bright, forward lean" described by the OP.
Well, I still have my HD220 that I rotate in and out of my stash of amps.
And yes the unit has become lean and mean and despite attempts to tube-roll using Amperex pinch waist 6922s and various 6H30s, the lean and mean signature remains.
I requested a schematic from Audio Research, but zero response despite emails and phone calls. Absolute ZERO.
I then took it upon myself to try to trace out the circuit.
The output stage is similar to if not the same as the SD135, whose schematics are easy to find online. My unit has the newer generation Thermaltraks and boy am I glad to discover that. The initial attempt was made to the output stage. I adjusted the bias drive to operate a wee bit higher current and the sound warmed up at the expense of being chesty and grainy.
Then I spent some time looking at the tubed gain stage. It's essentially a Diff-Amp gain stage using one plate of a 6922 directly coupled to paralleled 6H30, 230V B+ and with the existing tubes, plate V at coupling point was 115V which for a moment I thought was a nice midpoint value. But several things stand out. 6H30 cathode follower is at 115V against 23kOhms, which means the CF is extremely current-starved. The 6922 gain stage has a 38k plate resistor on each half to B+. A quick simulation of this circuit shows a moderate degree of 3rd order harmonics which explains the bright and lean sound. Power amps rarely measure poorly with a frequency range that is emphasized. But the human brain does interpret odd order harmonics as "edgy and lean". The saving grace of this model is that there are 2 trimpots for each channel for the tube stage, one controls the 6922 diff-amp's balance between the 2 triodes, one controls the bias. Where I found it (not sure if its the factory-desired plate V or a symptom of aging tubes) the grid bias is at 20V giving a plate V or 115. But simulation results tell me that in order to minimize 3rd order harmonics at a reference 1k sine wave I need a grid bias of 15V yielding a plate V of about 85. That essentially did the trick to remove the "edgy and lean" sound. Unless I have the schematics/service manual that describes reference voltages, I will never know whether the OP (as well as my own recent) experience is due to a gradual drift in the 6922s as they age, causing the tubes to run leaner and meaner, or it's audio research's voicing efforts right from the start. But one thing I can see from their older models (eg the D115), the gain stage 6922 runs a lot hotter with a distortion bias in favor of even order harmonics vs odd harmonics.
Bottom line is, this unit is still a keeper.